Skip to main content
Velor
All posts
PerspectiveApril 2026 · 5 min read

Resolution rate is the only AI support metric that matters

By Caden Boatright


AI support vendors love deflection rate. It's a compelling number — 70% deflection sounds like 70% of your tickets are handled. But deflection rate measures contact attempts, not outcomes. A bot that says “I'm sorry, I can't help with that — here's our FAQ page” counts as a deflection in most systems.

Resolution rate is harder to measure and harder to achieve, which is exactly why it's the right metric.

What deflection rate actually measures

Deflection rate counts conversations that didn't reach a human agent. That's it. A customer who got a wrong answer and gave up counts as deflected. A customer who got sent in circles and finally closed the tab counts as deflected. The metric can look good while the customer experience is bad.

What resolution rate measures

Resolution rate measures conversations where the customer's issue was actually resolved — without a human having to step in. To calculate it:

  • Resolved: The agent answered the question or completed the action, and the conversation ended without escalation.
  • Escalated: The agent handed off to a human — either because it hit a knowledge boundary, the customer asked for a person, or a guardrail triggered.
  • Failed: The agent gave a wrong answer, went in circles, or the customer abandoned after multiple unhelpful turns.

Resolution rate = Resolved ÷ (Resolved + Escalated + Failed)

Why resolution rate is harder — and more honest

To measure resolution rate accurately, you need to be willing to count failures. That requires honest evaluation methodology: running the agent on tickets with known outcomes, or collecting post-conversation signals (customer sentiment, whether they contacted support again, whether the action completed successfully).

It's also dependent on ticket mix. An agent handling routine, high-frequency questions — account lookups, policy questions, status checks — should achieve a high resolution rate, because the answer is almost always in the knowledge base. An agent handling complex technical questions or open-ended problems will have a lower natural ceiling.

What good looks like (by ticket category)

There's no universal benchmark because ticket difficulty varies so much, and we don't publish percentage ranges here: no customer dataset backs one yet, and a number on this page would be quoted as if it had. What to look for, category by category, when you run your own tickets through an eval:

  • Simple transactional (WISMO, password reset, order status): usually the highest-resolving category, because the answer is a lookup. If it is not, check that the agent can reach the system that holds the answer.
  • Billing questions (subscription tier, invoice lookup, proration): depends on how much of the billing system is exposed via API tooling.
  • Technical support (SDK questions, integration debugging): the most sensitive to KB quality and how current the docs are.
  • Policy questions (return policy, SLA terms): almost entirely dependent on KB completeness.

Read the rate per category, not just the blended number: the blend mostly tells you your ticket mix.

How to run your own baseline

The fastest way to understand what's achievable for your ticket mix: take your last 500 closed tickets, run them through the agent, and score each one pass or fail. This is what we do in Velor's free evaluation — and it's also something you can do with any AI support tool before you buy it.

The output you want: a breakdown by ticket category, not just an overall number. That breakdown tells you where to deploy first, what KB gaps to fill, and what human-in-the-loop patterns you still need.

Resolution rate won't be a perfect metric for every organization. But it asks the right question — did the customer get their problem solved? — and that's the question your customers are asking.

Numbers in this post that cite no source are rules of thumb, not Velor measurements. Velor has no customer deployment to measure yet.

Velor

Want to see how this applies to your support queue?

Send us your recent tickets — at least 300, ideally 500 — and get a resolution rate, before you commit to anything.

14-day free trial · no credit card